Merge pull request #212 from ceph/wip-4451

6 files changed, 58 insertions, 29 deletions

diff --git a/src/client/Client.cc b/src/client/Client.cc
index 420c3ad00f2..aae22ffa980 100644
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -1534,9 +1534,7 @@ void Client::handle_client_session(MClientSession *m)
   case CEPH_SESSION_OPEN:
     renew_caps(session);
     session->state = MetaSession::STATE_OPEN;
-    if (unmounting) {
-      _close_mds_session(session);
-    } else {
+    if (!unmounting) {
       connect_mds_targets(from);
     }
     signal_cond_list(session->waiting_for_open);
@@ -1966,6 +1964,8 @@ void Client::send_reconnect(MetaSession *session)
   resend_unsafe_requests(session);
 
   messenger->send_message(m, session->con);
+
+  mount_cond.Signal();
 }
 
 
@@ -3778,17 +3778,17 @@ void Client::unmount()
   }
 
   
-  // send session closes!
-  for (map<int,MetaSession*>::iterator p = mds_sessions.begin();
-       p != mds_sessions.end();
-       ++p) {
-    if (p->second->state != MetaSession::STATE_CLOSING) {
-      _close_mds_session(p->second);
+  while (!mds_sessions.empty()) {
+    // send session closes!
+    for (map<int,MetaSession*>::iterator p = mds_sessions.begin();
+	p != mds_sessions.end();
+	++p) {
+      if (p->second->state != MetaSession::STATE_CLOSING) {
+	_close_mds_session(p->second);
+      }
     }
-  }
 
-  // wait for sessions to close
-  while (!mds_sessions.empty()) {
+    // wait for sessions to close
     ldout(cct, 2) << "waiting for " << mds_sessions.size() << " mds sessions to close" << dendl;
     mount_cond.Wait(client_lock);
   }
diff --git a/src/mds/Capability.h b/src/mds/Capability.h
index 946afdc02b9..54d2312daeb 100644
--- a/src/mds/Capability.h
+++ b/src/mds/Capability.h
@@ -272,6 +272,7 @@ public:
   Export make_export() {
     return Export(_wanted, issued(), pending(), client_follows, mseq+1, last_issue_stamp);
   }
+  void rejoin_import() { mseq++; }
   void merge(Export& other) {
     // issued + pending
     int newpending = other.pending | pending();
diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index 3f090bb3238..3129ed7c267 100644
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -4097,20 +4097,25 @@ bool MDCache::parallel_fetch_traverse_dir(inodeno_t ino, filepath& path,
     frag_t fg = cur->pick_dirfrag(path[i]);
     CDir *dir = cur->get_or_open_dirfrag(this, fg);
     CDentry *dn = dir->lookup(path[i]);
-    CDentry::linkage_t *dnl = dn->get_linkage();
-    if (!dn || dnl->is_null()) {
-      if (!dir->is_complete()) {
-	// fetch dir
-	fetch_queue.insert(dir);
-	return false;
-      } else {
+    CDentry::linkage_t *dnl = dn ? dn->get_linkage() : NULL;
+
+    if (!dnl || dnl->is_null()) {
+      if (!dir->is_auth()) {
+	dout(10) << " not dirfrag auth " << *dir << dendl;
+	return true;
+      }
+      if (dnl || dir->is_complete()) {
 	// probably because the client created it and held a cap but it never committed
 	// to the journal, and the op hasn't replayed yet.
 	dout(5) << " dne (not created yet?) " << ino << " at " << path << dendl;
 	missing.insert(ino);
 	return true;
       }
+      // fetch dir
+      fetch_queue.insert(dir);
+      return false;
     }
+
     cur = dnl->get_inode();
     if (!cur) {
       assert(dnl->is_remote());
@@ -5041,8 +5046,32 @@ void MDCache::rejoin_import_cap(CInode *in, client_t client, ceph_mds_cap_reconn
 
   Capability *cap = in->reconnect_cap(client, icr, session);
 
-  if (frommds >= 0)
+  if (frommds >= 0) {
+    cap->rejoin_import();
     do_cap_import(session, in, cap);
+  }
+}
+
+void MDCache::export_remaining_imported_caps()
+{
+  dout(10) << "export_remaining_imported_caps" << dendl;
+
+  for (map<inodeno_t,map<client_t,map<int,ceph_mds_cap_reconnect> > >::iterator p = cap_imports.begin();
+       p != cap_imports.end();
+       ++p) {
+    for (map<client_t,map<int,ceph_mds_cap_reconnect> >::iterator q = p->second.begin();
+	q != p->second.end();
+	++q) {
+      Session *session = mds->sessionmap.get_session(entity_name_t::CLIENT(q->first.v));
+      if (session) {
+	// mark client caps stale.
+	MClientCaps *stale = new MClientCaps(CEPH_CAP_OP_EXPORT, p->first, 0, 0, 0);
+	mds->send_message_client_counted(stale, q->first);
+      }
+    }
+  }
+
+  cap_imports.clear();
 }
 
 void MDCache::try_reconnect_cap(CInode *in, Session *session)
diff --git a/src/mds/MDCache.h b/src/mds/MDCache.h
index 73780e26892..d837586a3ac 100644
--- a/src/mds/MDCache.h
+++ b/src/mds/MDCache.h
@@ -486,6 +486,7 @@ public:
   void rejoin_import_cap(CInode *in, client_t client, ceph_mds_cap_reconnect& icr, int frommds);
   void finish_snaprealm_reconnect(client_t client, SnapRealm *realm, snapid_t seq);
   void try_reconnect_cap(CInode *in, Session *session);
+  void export_remaining_imported_caps();
 
   // cap imports.  delayed snap parent opens.
   //  realm inode -> client -> cap inodes needing to split to this realm
diff --git a/src/mds/MDS.cc b/src/mds/MDS.cc
index 3b3b2d6dc2e..935fb0c417e 100644
--- a/src/mds/MDS.cc
+++ b/src/mds/MDS.cc
@@ -1504,6 +1504,7 @@ void MDS::active_start()
 
   mdcache->clean_open_file_lists();
   mdcache->scan_stray_dir();
+  mdcache->export_remaining_imported_caps();
   finish_contexts(g_ceph_context, waiting_for_replay);  // kick waiters
   finish_contexts(g_ceph_context, waiting_for_active);  // kick waiters
 }
diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index dc7ea23f763..11ab834d856 100644
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -574,7 +574,7 @@ void Server::handle_client_reconnect(MClientReconnect *m)
 
   // notify client of success with an OPEN
   mds->messenger->send_message(new MClientSession(CEPH_SESSION_OPEN), m->get_connection());
-    
+
   if (session->is_closed()) {
     dout(10) << " session is closed, will make best effort to reconnect " 
 	     << m->get_source_inst() << dendl;
@@ -636,15 +636,12 @@ void Server::handle_client_reconnect(MClientReconnect *m)
     }
       
     filepath path(p->second.path, (uint64_t)p->second.capinfo.pathbase);
-    if ((in && !in->is_auth()) ||
-	!mds->mdcache->path_is_mine(path)) {
+    if (in && !in->is_auth()) {
       // not mine.
       dout(0) << "non-auth " << p->first << " " << path
 	      << ", will pass off to authority" << dendl;
       
       // mark client caps stale.
-      inode_t fake_inode;
-      fake_inode.ino = p->first;
       MClientCaps *stale = new MClientCaps(CEPH_CAP_OP_EXPORT, p->first, 0, 0, 0);
       //stale->head.migrate_seq = 0; // FIXME ******
       mds->send_message_client_counted(stale, session);
@@ -652,11 +649,11 @@ void Server::handle_client_reconnect(MClientReconnect *m)
       // add to cap export list.
       mdcache->rejoin_export_caps(p->first, from, p->second);
     } else {
-      // mine.  fetch later.
+      // don't know if the inode is mine
       dout(0) << "missing " << p->first << " " << path
-	      << " (mine), will load later" << dendl;
-      mdcache->rejoin_recovered_caps(p->first, from, p->second, 
-				     -1);  // "from" me.
+	      << " will load or export later" << dendl;
+      mdcache->rejoin_recovered_caps(p->first, from, p->second, -1);
+      mdcache->rejoin_export_caps(p->first, from, p->second);
     }
   }